Research on Personalized Recommendation Strategy for Teaching Content of Sports Culture Based on Deep Learning

The user layer is the layer that directly interacts with learners and displays sports culture teaching content recommendation results. Through this layer, the platform provides corresponding services to learners. In this layer, learners can learn, collect, and evaluate sports culture teaching content, communicate with other learners, and record learning notes, etc. This layer not only displays personalized sports culture teaching content to learners, but also provides services for their learning process. The user layer is the layer with the closest intuitive feeling of the whole platform and the learners, through which the users feel the quality of the recommended sports culture teaching content and the rationality of the function module settings, etc. Therefore, this layer affects the evaluation of the platform by learners.

The business layer is the core layer for providing personalized sports and cultural teaching content for learners, including recording learners’ learning behaviors, such as learning time, resource clicks, learning conditions, etc., as well as processing the data accordingly and training the neural network model to make corresponding predictions so as to personalize the recommended sports and cultural teaching content. The business layer is not only the core execution layer of the recommendation model mentioned in the previous section, which is the guarantee of personalized sports and cultural teaching content recommendation service, but also the core layer of the whole platform, which is the most important logical processing layer of the platform to ensure the stable operation of the whole platform, which directly affects the learners’ experience and satisfaction.

The storage layer stores the corresponding data through the corresponding database, mainly storing user information, physical culture teaching content information, learner learning records, learners’ study notes and so on. The purpose of the storage layer is to provide data support and services for other layers, while ensuring the integrity, security, and reliability of the data. Personalized sports and culture teaching content recommendation service cannot be separated from the support of data, the storage layer is the foundation of personalized sports and culture teaching content recommendation service, and is also the basis of personalized sports and culture teaching content recommendation service. Data is an important guarantee for the effective operation of the whole platform, so it is necessary to ensure the comprehensiveness and rationality of the database design as well as the stability and security of the storage layer.

The management layer is the layer that directly interacts with the administrator, and mainly manages the platform-related contents, such as user information management, addition, deletion, checking and modification of physical culture teaching contents, chat record management, question management, etc. The management layer is the layer that directly interacts with the administrator. The management layer is mainly oriented to the platform administrator, and its core function is the management of data, including data statistics, data analysis, etc. The most important thing for the management layer is to ensure that the sports culture and teaching content are managed. The most important thing for the management is to guarantee the richness and reliability of the sports and culture teaching content, guarantee the legality and reasonableness of the users, chat records and other contents, and deal with the users’ opinions and suggestions in a timely manner, and the quality of the management also affects the users’ satisfaction with the platform.

The attributes of learners and sport culture teaching content on different learning platforms are diversified, and there are many factors affecting learners’ selection of sport culture teaching content, which may include gender specialty, learning goal, content preference, learning style, cognitive level, motivation and other characteristics. On the other hand, physical culture teaching content may have intrinsic attributes such as resource style and interaction mode, so it is necessary to find the association between learners and resources among many features and establish a feature selection model to complete the input process of the recommendation method.

In the MIFS-based feature selection method, the information metric evaluation function is crucial, although the functions are in various forms, the purpose is to select the subset of features that have the greatest relevance to the category. The evaluation function of generalized information metrics can be expressed as: (1)J(f)=α⋅g(C,f,S)−δ

Where S for the selected features, f for the candidate features, C for the category, function g(C,f,S) is the amount of information between C, f, S, that is, the selected features to add candidate features and the degree of correlation between the category, α is the adjustment coefficient, used to regulate the degree of information brought by the addition of f, δ is the penalty factor, used for the degree of redundancy brought by f to S. Therefore the simplest and most intuitive information metric evaluation function can be: (2)J(f)=g(C,f)

Where g() is the mutual information, β is the adjustment coefficient, at this time directly all the candidate features to calculate the evaluation function, in order to prevent too much redundancy in the pre-processing, considering the correlation between the features, through the correlation between feature f and a single feature s has been selected on the f levy penalties, the final evaluation function can be expressed as: (3)J(f)=g(C,f)−β∑s∈Sg(s,f)

Determining learners’ preferences for content features for teaching physical culture is particularly important, and the features that have been selected indicate the identification of a number of features that will influence learners’ choice of resources, such as the knowledge content of the resources and the duration of learning.

Learners’ multiple learning of a physical culture teaching content or certain physical culture teaching content can be of great help in understanding learners’ behaviors and what they are interested in. In this paper, we propose a two-part graphical association model of learner-resource, defining the set of learners as: (4)L={l1,l2,l3…l4}

Resource pooling for: (5)R={R1,R2,R3…Rn}

Therefore, a binary relationship matrix X_m × n can be obtained which consists of the set of learners and the set of resources, the row vector represents the learners, and the column vector represents that the learners have learned the resource, if X_m × n is equal to 1, it means that the learners have learnt the resource, and if X_m × n is equal to 0, they have not learnt it.

The learners’ frequency of learning the content of physical culture teaching can reflect different preferences, and the average frequency of resource learning can be defined as: (6)r(li)¯=∑j=1ik(li,rj)n(li)

In the problem of sports culture teaching content recommendation, classical machine learning algorithms can be effectively trained on historical learning data, but traditional machine learning algorithms or simple neural network models often can not meet the actual demand, can not guarantee convergence to an optimal solution, the proposed method in this paper designed a deep neural network model (S-CNN), the output to determine whether the learner to learn a certain physical culture teaching content or the importance of that physical culture teaching content, so a deep neural network model is designed to solve the problem. The key to applying deep neural networks to the sports culture teaching content recommendation scenario lies in modeling the learners’ historical learning records and mining the implicit features of the original data, so as to standardize the input and output layers when constructing the training model, and the feature selection model based on the MIFS and the learner-resource bipartite graph correlation model mentioned above effectively solved the input and output parts of the deep neural network.

The specific design of the deep neural network model is as follows: 1)

Hidden layer design: the hidden layer adopts the sigmoid activation function of the sigmoid system, and its functional expression is shown in Equation (7), which is in the shape of S in the two-dimensional coordinate axes, x is the value of the horizontal coordinate, and f(x) is the value of the vertical coordinate: (7)f(x)=11+e−x

1)

Classification evaluation metrics

Whether a classification model predicts correctly usually involves the following terms:

Checking accuracy (P), recall (R) and F1-score value (F) is usually the higher the better, indicating that the probability of the recommended physical culture teaching content being learned by the learner is greater, i.e., the recommended results are more in line with the needs of the learner, and their mathematical formulas are shown in Equation (9): (9){ P=TPTP+FP R=TPTP+FN F=2×P×RP+R

where n denotes the sample size, X_i denotes the ird actual value, and Y_I denotes the ith predicted value.

In order to verify whether the deep neural network-based physical culture teaching content recommendation algorithm meets learners’ needs, a crawler program was used to crawl the free courses and user-related data in an online learning website. After cleaning, the dataset includes 730 courses, 1,320 course-related physical culture teaching content profiles, and 2,140 users with learning records. Here, “sports culture teaching content profiles” refers to course-centered content, including course video profiles, test bank profiles, discussion group profiles, etc. The content profiles of a particular sport culture’s teaching content profile include course video profiles, test bank profiles, discussion group profiles, and so on. The content profiles not only contain their own profile information, but also include an overview of the course to which the resource belongs.

In order to ensure the superior quality of learners’ viewing history data, the viewing history data of course resources needs to be thoroughly pre-processed. After analyzing the data, it was found that about 21% of the viewings were less than two minutes long, while the largest proportion of the data were viewed between 2 and 20 minutes. Short viewings may be due to learners mis-touching or disinterest in the physical culture pedagogical content. Therefore, we retained the learning history for more than two minutes.

The model loss function in the paper is iteratively optimized using stochastic gradient descent method. In order to compare the performance of different recommendation algorithms, when the number of recommendations N is 5, 10, 15 and 20 respectively, the checking rate, recall rate and F1 value of four methods, namely SVD, collaborative filtering, LSTM and deep neural network, are calculated for comparison. Among them, the SGD learning rate is set to be randomly selected between 0.001, 0.005, and 0.01 to find the optimal learning rate. The comparison results are shown in Fig. 2, when the number N of recommended physical culture teaching contents is small, the collaborative filtering method, LSTM method and deep neural network (about 40% of the checking rate) method have high checking rate, and with the increase of N, the SVD method also achieves better results. In terms of recall metrics, LSTM methods and deep neural network methods are outstanding. From the analysis of F1 values, it can be concluded that the deep neural network method proposed in this paper can achieve better results at N=15 and N=20.

Figure 2.

Check the accuracy, recall rate and f1 value

Synthesizing the research results of academic theories in recent years, sports culture in colleges and universities refers to the sum of material and spiritual achievements created in the process of sports teaching, research and management in the specific atmosphere and environment of colleges and universities, with students as the main body, teachers as the leading role, and with the goal of promoting the all-around development of college students, and with physical exercise as the means and all kinds of sports knowledge as the main content. It is a product of the organic combination of campus culture and sports culture in today’s society, and belongs to a very complicated and special subculture form.

China’s sports culture resource guarantee system adopts decentralized construction, unified storage, and unified retrieval for the integration of resources. The system is built with a sports culture content featured resource integration platform to integrate and summarize the sports featured resources dispersed in various sports colleges and universities, and establishes a unified entry format for each type of resource, constructs a bottom-level sub-type resource database according to the category, and on top of the bottom-level database, the resources can be extracted according to the needs of each type of resource. According to the need to extract the corresponding conditions of various types of resources into a thematic database, such as the track and field thematic database and preparation for the Winter Olympic Games thematic database. Research on the concept and characteristics of sports information resources is very limited. By the end of 2023, China’s sports culture teaching resources have collected nearly one million pieces of data, covering sports experts and scholars, sports books, journal papers, dissertations, sports videos, sports personalities, and other types, and metadata citation is performed separately for different types, and the specific construction situation is shown in Table 1.

The metadata citation of the above types of resources is the same as that of the citation fields of various general information resources, but there are also four additional fields for sports characteristics, which are divided into: sports (including Olympic sports, leisure activities, traditional sports, etc.), sports sciences (including: sports ergonomics, sports psychology, humanities and sociology of sports, traditional national sports, sports training, sports engineering, sports information, sports statistics, sports art, school sports, etc.), sports events (including: world-class sports events, intercontinental sports events, domestic sports events, foreign other sports events, etc.), sports organizations (including: international sports events, intercontinental sports events, domestic sports events, foreign other sports events, etc.) Statistics, Sports Art, School Physical Education, etc.), Sports Events (including: world-class sports events, intercontinental sports events, domestic sports events, other foreign sports events, etc.), Sports Organizations (including: International Sports Organizations, Domestic Sports Organizations), and so on.

The use of word cloud diagrams for physical education courses in various branches of the number of people to choose courses to show, you can see which course is more popular or popular with students, in the subsequent development of the curriculum plan can be adjusted based on the information in the chart. Qualitative description of the course recommendation results for a student as shown in Figure 3, the student for science and engineering students, where the actual course for the students to actually choose the course a total of 94 courses, recommended courses selected N for 32, a total of 32 recommended courses, recommended subjects include both basic courses, including professional courses, recommended courses and the actual course compared to the recommended correctness rate of 0.75, the majority of recommended courses for engineering courses, but also recommended courses for engineering courses, as well as professional courses. Courses for engineering courses also include popular basic courses that are recommended but not necessarily selected in subsequent studies, which can also open up new possibilities and provide students with personalized learning programs.

Figure 3.

Recommended results

In order to verify the feasibility of the personalized recommendation system for deep learning sports culture teaching, this study is based on the monitoring and analysis of college students’ sports literacy in a college in X city, 390 students were randomly selected from a public college within X city, and the teaching guidance supported by the personalized recommendation system will be implemented for this group of students in their daily physical education classes. The physical education performance before and after one semester of instruction was quizzed. The results are shown in Table 2. It is worth noting that this study found that the personalized recommendation system had a more significant effect on the enhancement of college students’ sports knowledge and sports awareness (P<0.01), and that the subjects’ sports behaviors (time spent on medium- and high-intensity exercise) increased significantly (P<0.01). Sports knowledge contains knowledge of sports nutrition, health care, safety and other knowledge, and its improvement needs to be taught and explained through the corresponding knowledge, relying on personalized scientific guidance can not significantly improve the comprehensive performance of sports knowledge. Sports awareness is the awareness of participating in sports activities independently, which can be continuously enhanced in sports.

Exercise guidance supported by the deep learning-based personalized recommendation model of physical culture teaching content has a greater improvement on college students’ physical fitness level scores, and the results are shown in Table 3. Among them, the improvement of BMI and 1000-meter running performance was most obvious and there was a significant difference (P<0.01). Lung capacity, 50-meter run, and standing long jump scores had some improvement, and there was a significant difference (P<0.05). Seated body flexion and pull-up scores were improved but not significantly different (P>0.05). Seated forward flexion represents hip flexibility, overweight obese high school students due to the accumulation of fat in the body caused by the hip joint activity is limited, 8 weeks of personalized scientific guidance can be improved, but the significant improvement of flexibility still need more training cycles. In addition, the students’ pull-up performance was not significantly improved, probably due to the fact that the overweight and obese students’ own body weight was too large, resulting in relatively small strength, and the improvement of performance needs to reduce body weight while strengthening their own strength. Overall, the exercise guidance program supported by the personalized recommendation system has achieved the “right remedy”. At the same time, students were alerted to the importance of exercise, and thus were constantly urged to exercise.